AI transparency
The characters in Vefraia are AI. This page explains what the AI does, what it doesn’t do and what data it sees (EU AI Act, Art. 50).
Last updated: September 2026What the AI does
Every character has a profile based on four communication models. After each contribution a language model estimates what lands with the characters, who speaks next and where the scene tips. In “Learn” mode it then explains why.
How you recognise AI
Characters, inner voice and insights are always AI-generated. A notice appears at the start of every scenario and permanently in the header.
What data the AI sees
Only what happens in the scenario: chosen options and your own words. You are told once the first time you type your own words. Processing takes place in the EU.
What the AI doesn’t do
It doesn’t judge people – insights describe the conversation, never the person. It is not used for HR decisions. It doesn’t recognise emotions from voice or face; Vefraia is text-based. It creates no individual reports for employers – only aggregates for five or more people.
Models and providers
We use language models from Anthropic, PBC (USA) via the Anthropic API. Which model a scenario uses is decided after quality testing. The list of sub-processors is provided to organisations in the DPA.
Limits
The simulation is a probability, not a prediction. Real people react differently. Vefraia does not replace counselling, therapy or mediation.
Contact
Questions about the AI: web@cyan-it.de